The addition of the FLE subsystem to the shell pulls in a very large swath of query code. Some of this is clearly necessary, as FLE needs to understand more about queries than it did in the past.
However, other parts seem like they should't be coming along for the ride, like the query planner or the sort subsystem.
The inclusion of this extra code causes a roughly 25% startup penalty for a dynamically linked shell, which given how many shells are started in the test suites adds up to a non-zero amount of delay.
Some effort should be made to decompose the query code into the interfaces and implementations that are only necessary for to support FLE so that these extra dependencies can be pruned.